Databricks Machine Learning Professional Question 17
Select 3A data scientist is working on a machine learning project using Databricks Feature Store. They need to create a feature table that will be used across multiple projects, but they also want to ensure that the feature table is updated with new data periodically. Additionally, they want to track and compare experiment results while iterating on machine learning models. Which of the following steps should they take to achieve this workflow?
- A
Use the Databricks Feature Store API to create a feature table and periodically overwrite it with an updated DataFrame.
- B
Merge new feature data into the existing feature table using the Databricks Feature Store API.
- C
Log experiment parameters, metrics, and artifacts using MLflow's tracking capabilities.
- D
Use the Databricks Feature Store UI to schedule automated feature table updates.
- E
Directly train the model on raw data without using a feature table to simplify the workflow.
Show answer and explanation
Correct answers: A, B, C
Explanation
To achieve the described workflow, the data scientist needs to leverage the Databricks Feature Store API for creating, updating, and merging feature tables, ensuring features are managed and reusable. Additionally, they need to use MLflow for experiment tracking to log and compare results. Directly training on raw data or relying solely on the Feature Store UI does not align with the workflow requirements.
- A. Correct.
This is correct because the Databricks Feature Store API allows users to create and overwrite feature tables, which is necessary for ensuring the table is up-to-date with new data.
- B. Correct.
This is correct because merging new data into existing feature tables is a common practice in Databricks Feature Store to keep the table consistent and avoid complete overwrites when only partial updates are needed.
- C. Correct.
This is correct because MLflow's tracking capabilities are designed to log experiment parameters, metrics, and artifacts, enabling the data scientist to track and compare experiment results effectively.
- D. Incorrect.
This is incorrect because, while the Databricks Feature Store UI offers some functionality, it does not currently have built-in support for scheduling automated updates for feature tables. This task requires custom scheduling and scripting.
- E. Incorrect.
This is incorrect because directly training on raw data bypasses the use of the Feature Store, which is central to this workflow. Feature tables are crucial for feature management and reusability across projects.